Thanks. I was more curious about how non-Wikipedia sources (e.g. news articles) are judged to be reliable sources from an automation standpoint. The paper makes reference to https://en.wikipedia.org/wiki/Wikipedia:Reliable_sources, but how does the LLM implement this guidance?
In the paper they only talk about Wikipedia, as it's the main purpose of the paper. However, in their github repo they say that you can use it to generate any article with citations. https://github.com/stanford-oval/storm
There is no solid solution to judge if a reference is reliable or not. You simply need to take care of this part on your own by implementing your custom method.
Quick follow up. I am now exploring STORMS as essentially a search engine for references on topics of interest. STORMS appears to have access to publications that are normally blocked by a pay wall. Here is an example https://www.stc.org/techcomm/2017/05/10/designing-for-a-culturally-inclusive-democracy-a-case-study-of-voter-registration-outreach-postcards-in-latino-communities/ What service do you use to get access?
Thanks for this great work. Can you say more about how you identified "trusted" sources?
The paper explains that they use other pages from Wikipedia that are similar to the same topic of this new article as trusted sources.
Thanks. I was more curious about how non-Wikipedia sources (e.g. news articles) are judged to be reliable sources from an automation standpoint. The paper makes reference to https://en.wikipedia.org/wiki/Wikipedia:Reliable_sources, but how does the LLM implement this guidance?
In the paper they only talk about Wikipedia, as it's the main purpose of the paper. However, in their github repo they say that you can use it to generate any article with citations. https://github.com/stanford-oval/storm
There is no solid solution to judge if a reference is reliable or not. You simply need to take care of this part on your own by implementing your custom method.